Files in this item



application/pdfTONG-THESIS-2016.pdf (510kB)
(no description provided)PDF


Title:DPLearn: an effective but concise learning framework based on discriminative patterns
Author(s):Tong, Wenzhu
Advisor(s):Han, Jiawei
Department / Program:Computer Science
Discipline:Computer Science
Degree Granting Institution:University of Illinois at Urbana-Champaign
discriminative patterns
Abstract:Pattern-based classification was originally proposed to improve the accuracy using selected frequent patterns, where many efforts were paid to prune a huge number of non-discriminative frequent patterns. On the other hand, tree-based models have shown strong abilities on many learning tasks since they can easily build high-order interactions between different features and also handle both numerical and categorical features as well as high dimensional features. By taking the advantage of both modeling methodologies, a natural and effective way is proposed to resolve pattern-based learning tasks by adopting discriminative patterns which are the prefix paths from root to nodes in tree-based models (e.g., random forest). Moreover, the number of discriminative patterns is further compressed by selecting the most effective pattern combinations that fit into a generalized linear model. Note that this method is a general framework, which is applicable on both classification and regression tasks by using different loss functions. Extensive experiments demonstrate that the discriminative pattern-based learning framework (DPLearn) could perform as good as previous state-of-the-art algorithms, provide great interpretability by utilizing only very limited number of discriminative patterns, and predict new data extremely fast. More specifically, in classification tasks, DPLearn could gain even better accuracy by only using top-20 discriminative patterns, while in regression tasks DPLearn delivers reasonable performance that is comparable to complex models, which shows that framework so generated is very concise and highly explanatory to human experts.
Issue Date:2016-04-20
Rights Information:Copyright 2016 Wenzhu Tong
Date Available in IDEALS:2016-07-07
Date Deposited:2016-05

This item appears in the following Collection(s)

Item Statistics